Algorithm For Multi-sensors Data Fusion In WSN And Its Application In Forest Fire Prevention

In recent years, with the development of the theory of wireless sensornetworks, it has been widely used in forest fire monitoring. Wildfire is a highlydestructive disaster that spread quickly. The use of advanced technological toachieve an early warning of wildfire is essential for the protection of wildresources. Compared with traditional means of forest fire monitoring such asartificial patrol, watchtowers observation, airplane cruise and satellite remotesensing etc., wireless sensor network has many advantages such as low cost,comprehensive perception, high accuracy. And many scholars have conductedthe research on the application of wireless sensor network in the early wildfirewarning system.However, a WSN is usually constrained by various resources, includingbattery capacity, computational power, data storage, and communicationbandwidth. The effective collection, integration, and transfer of the measureddata are important factors for the success of WSNs. Aiming at the problems ofthe WSN，data fusion is a multilevel, multifaceted process dealing with theautomatic detection, association, correlation, estimation, and combination ofdata and information from multiple sources. By combining the data collected by multiple sensors at different times and in different places, data fusion canprovides more accurate and comprehensive information about the area ofinterest than a single sensor can offer.In this paper, aiming at the existing problems in the application ofwireless sensor networks in forest fire monitoring, we proposes a hierarchicalclustering data fusion algorithm, to realize the WSN energy saving, real-timeand reliability requirements in the field of fire warning applications. In-clustersensor node uses the adaptive weighted fusion algorithm (AWFA) achievedata-level fusion of the raw data, reduce the redundancy of raw data thusreduce the traffic from the in-cluster sensor nodes to the cluster head node.Then, we establish a D-S evidence theory based recognition framework in thecluster head node, through the decision level fusion of the feedback signal ofthe cluster members, can get the global optimal estimation of the fire, and therecognition accuracy of fire event and robustness of the network has beenimproved.At the same time, we also put forward some fire warning auxiliaryalgorithm, including heterogeneous sensor data homogenization methods,judgment algorithm of numerical errors of sensor, and evidence conflictsolution of Dempster-Shafer evidence theory. The experimental results showthat the algorithm can not only to achieve the real time processing of sensorydata on the same time but different spatial, but also can effectively eliminatethe redundant data, save the network bandwidth and energy, improve the recognition accuracy of target event and robustness of the network, and it canwork correctly in the case that the number of faulty nodes do not exceed40%of the total number of nodes. Thereby improving the scientificdecision-making of the wildfire monitoring system, and ensure the timelinessand accuracy of the wildfire early warning.